# AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society

*Source* · `src-piao-2025-agentsociety` · status: **n/a**

| field | value |
|---|---|
| kind | Preprint |
| authors | Jinghua Piao, Yuwei Yan, Jun Zhang, et al. (Tsinghua University) |
| venue | arXiv |
| date | 2025-02-12 |
| url | https://arxiv.org/abs/2502.08691 |
| arxiv | 2502.08691 |
| accessed | 2026-10-24 |

## Notes

Over 10k agents and about 5 million interactions; agents with profiles, emotion, needs, and cognition layers; urban, social, and economic spaces (firms, banks, government); Ray actors, asyncio, MQTT bus; DeepSeek API chosen to avoid rate limits. Experiments on polarization, inflammatory messages, universal basic income, hurricane shock.

## Backlinks (computed)

**cited by:** [Many-agent LLM societies](../technique/llm-agent-societies.md)

## Citation

Jinghua Piao, Yuwei Yan, Jun Zhang, et al. (Tsinghua University) (2025-02-12). *AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society*. arXiv https://arxiv.org/abs/2502.08691

---
Provenance: created in run `2026-10-24-techniques-llm-agents`.